future hardware costs do not erase losses on existing capex expenditures, if they bought an (overpriced) nvidia GPU and then it turns out local LLMs or a Chinese competitor can do it for much cheaper investors effectively notice the mortgage is underwater. Tech getting cheaper is only handy if your company (e.g. ChatGPT) didn't already make a big gamble they can't sell off (for fear of hurting the cost of the asset they're trying to sell) see also coin "reserves".

Just because they’re slashing they’re prices while they compete for users doesn’t mean the cost of inference came down at the same rate or at all.